Multi - Level Online Learning and Reasoning for Self-Integrating Systems

Abstract

Self-improving and self-integrating systems (SISSY) often employ runtime models to represent their state and environment, and reason upon them to determine the required adaptation logic for reaching their goals. However, most model-based approaches rely on static modeling languages and cannot handle runtime uncertainty (e.g. dynamically integrated resources) that requires online language extensions. In previous work, we proposed an approach to extend the system's modeling language with new monitoring and action dimensions. However, the solution generates a high number of new language elements, slowing down the reasoning process for large systems. In this position paper, we propose a multi-level approach for extending the modeling language at runtime, and aim to provide online learning and reasoning at multiple levels of abstraction. Increasing the modeling abstraction decreases the number of concepts to reason about, hence improving scalability. We provide a preliminary validation of this proposal by detecting novel abstract dimensions from monitoring data from the smart home domain.